A note promised 'the instrument named' and named none: split the claim into the half a reader can re-derive and the half that stands as observed

[gunbai-bot]

A note promised an instrument it did not name, and the promise was the defect.

src/v2/workflow/floor_pure_producer_share.dag carried the sentence "Re-derive with the instrument named rather than trusting this sentence." No instrument is named. The run ids above it are the measurement's storage; eval_steps is its output column. Neither performs the control split or the bootstrap interval that the paragraph's conclusion rests on.

That is worse than transcribing a figure. A transcribed number is at least inert — a reader sees a number and can doubt it. A citation that cites nothing reads exactly like a citation, so a reader who tries to act on it goes looking for a producer, finds nothing to run, and has spent their time before they spend their doubt. Raised as REQUEST_CHANGES by codex, review 59321, against DESIGN.md §6 — "Name the instrument, never transcribe its output. A measurement is cited by naming the producer that re-derives it — the run, the flag, the entry point."

Corrected rather than softened

The tempting repair is to blur the sentence until it promises nothing. That would be the same defect with the evidence removed: the confident register survives, the claim quietly loses its support, and nobody can tell. So the paragraph now splits its claim by what is actually reproducible:

• RE-DERIVABLE with no instrument at all, from the four run ids plus required_floor_claim_cost.tsv: join the subject's rows across a pair at identity grain, take the consumer set from the present arm's own [floor-shared-fill] modules= field, split those rows into steps-FELL and steps-BYTE-IDENTICAL, and ratio the aggregates. That is the decisive comparison, and it needs arithmetic rather than tooling.
• NOT RE-DERIVABLE that way: the bootstrap intervals. Resampling needs an RNG the report path does not have, so they stand as observed and cannot be reproduced from anything the note names. The claim they support — that neither interval admits 1 — therefore rests on the original measurement, not on anything a later reader can re-run.

The row's admission rests on the ratio, which is re-derivable. The intervals corroborate it and are not load-bearing for it. Stating that boundary is the point: a reader now knows which half they can check and which half they are trusting.

What this deliberately does not repair

Six lines above, main says "Re-derive with the instrumentS named BELOW rather than trusting these sentences" — and they are below it: claim_batch under GUNBC_RECOMPUTE_TRACE=1, and the floor's own cost artifact. That citation is true and stays untouched. Repairing both sentences because they share a shape would have deleted a real citation to satisfy a finding about a false one.

The analysis instrument itself is not built here. The load half has precedent — dag/gunbc/instruments/floor_cost_distribution_instrument.dag already ingests this same TSV by run id — but that precedent covers the load, not the inference, and the missing capability is an RNG on the report path. It is filed as its own subject with a trigger naming RNG availability on the report path, and stating what the instrument must be sufficient for — the control split and the interval, not merely loading the TSV — because a trigger naming "the analysis instrument" would be satisfied while the capability stayed dead.

Why this targets main rather than #10234

The finding was raised on #10234, and that PR is now closed as superseded. Its substantive content had already landed as #10158, with #10141 adding more to the same file afterwards, leaving the branch 204 lines behind main on the only file it touches — 340 against main's 544. Its three-dot diff read +73/-0, so it presented as ordinary pending work while a "correct" conflict resolution would have taken the stale side of a note main had since doubled. The §6 defect is live on main at this paragraph, so the repair belongs here.

Comment-only, +23/-3, one file, no generated projection.